259. VARIMA

Vector ARIMA

Differencing applied component-wise to each of the series.
In practice, cointegrated systems use VECM (Vector Error Correction) rather than VARIMA — VECM models the long-run equilibrium directly.

Parameters: , , ,
Orders: , , ,

Example:

Given

  • Orders: , , ,
  • Coefficient matrices:

  • Intercept:
  • Initial conditions: ,
  • Two series stacked into :
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
12 10 8 11 14 12 9 13 16 14 11 15 18 16 13 17
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
8 9 7 6 10 11 9 8 12 13 11 10 14 15 13 12

Step 1 — formula

Substitute orders into the VARIMA recursion:

Difference — apply component-wise:

In differenced space, the model is VARMA:

Forecast the difference (set ):

Undifference — convert back to a forecast for :

Innovation:

Step 2 — apply at

First difference: .

Plug in , , , :

Step 3 — iterate

Pipeline at each : difference VARMA-forecast undifference. Values rounded to 4 decimal places.

2
3
4
5
6
7
8
9
10
11
12
13
14
15
16